Velocity 09: Frank Mashraqi, "Best I/O is No I/O"
When overnight growth happens (such as when your site gets mentioned on a high traffic site such as Slashdot), it can be overwhelming. Especially because you must take immediate steps to accommodate the surge in traffic. While the front-end is relatively easy to scale-out, it’s squeezing more performance out of the back-end that is often the source of headaches.
For MySQL driven websites, horizontal sharding is often the long term solution to scale. But what if you cannot afford to stop your site and work on implementing the sharding layer? Is there any hope for you? Luckily, yes! You can perform quick analysis of your data access patterns and take actions to ensure your working data set fits in the memory. And, in the cases where disk I/O cannot be avoided, you can ensure that your data is laid out on disk in a manner than allows you to retrieve it with minimal number of disk seeks. Most of the strategies discussed in this session, require little or no application changes, allowing you to improve performance literally overnight.
We will explore strategies on how to optimize your data based on application access patterns and make your working data set fit within memory to decrease I/O and response times.
Frank Mashraqi is a renowned speaker and scalability advisor to several startups. He comes to NetEdge with nearly a decade of scalability, engineering management and monetization experience. Prior to NetEdge, Frank was Director of Business Operations and Technical Strategy for Fotolog where he played a pivotal role in helping Fotolog scale to become the 13th largest website on the Internet (Alexa: based on traffic) and the third most actively used social network in the world (ComScore). Frank holds a BBA in Accounting and a BS in Computer Information Systems.